1d ago

Flaw in Person-Δ Metric Overstates Progress in Single-Cell Perturbation Models

——0——
Original post
Anshul Kundaje#1675@ANSHULKUNDAJEOPRafael IrizarryRafael Irizarry@RAFALAB

A flaw in Person-Δ may be overstating progress in single-cell perturbation prediction models. Pearson warned about this in the 19th century: reusing the same controls induces spurious correlation. Split the controls, and much of the claimed prediction power fades. Link below 👇

7:45 AM · May 18, 2026 View on X
QUOTE POSTAnshul Kundaje#1675Anshul Kundaje@ANSHULKUNDAJE

Before building and scaling "AI virtual cells" or ML models in bio, think about the properties of the data and deeply consider whether you are optimizing the right measures. This is the "Bitter Lesson" of AI for Bio.

Rafael IrizarryRafael Irizarry@rafalab

A flaw in Person-Δ may be overstating progress in single-cell perturbation prediction models. Pearson warned about this in the 19th century: reusing the same controls induces spurious correlation. Split the controls, and much of the claimed prediction power fades. Link below 👇

2:45 PM · May 18, 2026 · 19.4K Views
4:50 PM · May 18, 2026 · 14.3K Views